Right now DictVectorizer is unable to handle features that have more than one value per key, such as: {'a': ['b','c'], 'd': 'e'} It would be great to simply encode a=b and a=c as the names of these features and 'just work.'